Exploring the Application of Artificial Intelligence in Design Courses in Colleges and Universities: an Example from Graphic Design

In the Internet and information age, artificial intelligence technology is reshaping the working mode of various industries with its powerful data processing ability and intelligent operation characteristics [1-2]. In the field of education, especially in the cultivation of innovation ability and practical skills, the introduction of AI technology has brought great changes to the traditional teaching mode [3-4].

The goal of design training is to strengthen students’ creativity and imagination through their theoretical knowledge, such as spatial design, design history and theory, general art, etc., and to combine graphic symbols with design methods to form a specific program from planar to spatial design [5-7]. Artificial intelligence as an auxiliary tool, the use of expanding space and increasing design resources as a data background, to meet the problem of students in the design can not be practically operated, to help teachers to integrate technology and education, through cross-disciplinary to provide a richer data support, for the design of the more complex problems arising in the design to get a rapid solution and optimize the available solutions [8-11]. Improve students’ subjective initiative. Artificial intelligence lays a data foundation for the cultivation of comprehensive and high-quality talents, constituting a perfect combination of design disciplines, educational disciplines and technology, so that design teaching keeps pace with the times [12-14].

As a design discipline that emphasizes the combination of creativity and practicality, the integration of artificial intelligence technology into its teaching process can not only provide richer and more diverse learning resources, but also create a more personalized and interactive learning environment [15-16]. Currently, design education is facing the double challenge of cultivating students’ innovative thinking and practical ability. In this context, exploring effective strategies for AI-enabled design professional curriculum teaching has become a common concern for educators. Through the application of intelligent teaching tools, students’ learning interest can be effectively stimulated, their creative inspiration can be enhanced, and then the overall development of students in design thinking and skills can be promoted [17-18]. Educators and scholars expect that through the help of artificial intelligence technology, design education will be more in line with the needs of the times, and cultivate high-quality talents who have a solid foundation in design theory and are also able to flexibly utilize modern technology to carry out innovative design [19-20].

Literature [21] attempts to compare the design discipline with science, humanities and other disciplines, aiming to change the design profession’s traditional profession-oriented design education goals to emphasize the intrinsic value of design, and the analysis promotes the progress of design professional development. Literature [22] reveals the lack of teaching of key aspects of viewpoint and process in the design process of designers, and puts forward an optimization plan to promote the improvement and optimization of design professional teaching. Literature [23] explored the design professional teaching effect empowered by additive technology, based on mixed research method analysis found that the introduction of additive technology into the design course teaching was once recognized by teachers and students, and it has great potential in design course teaching. Literature [24] builds a framework for teaching cross-cultural corporate identity graphic design courses, which helps students to understand and apply local culture and integrate into graphic design work. Literature [25] introduces the core of graphic design technology, its application and related skill competitions, and argues that the world standard of graphic design technology needs to be introduced into graphic design classroom teaching in order to realize the internationalization of graphic design teaching and graphic design work. Literature [26] discusses the need to cultivate not only professional design skills, but also critical thinking and a sense of ethical behavior in graphic design education in colleges and universities, which not only meets the commercial needs of social design, but also maintains social norms through its own ethical awareness. Designers continue to contribute their knowledge and talents to the creation of social art and aesthetics, and play an important role in more dimensions. Scholars focus on the theme of designer training, based on the intrinsic value of design, the contribution of social moral norms, and the enhancement of designers’ professional skills to explore the optimization and improvement of the design professional teaching program, which contributes to the promotion of the designer talent training mode and method.

Literature [27] examined the research related to the practice of AI-enabled education field from 1970-2020, identified the main research themes, AI personalized learning, potential applications of deep learning machine learning algorithms in the learning process, AI interactions in the educational process, intelligent educational data analysis, etc., and emphasized the ethical issues in the process of AI-enabled education. Literature [28] uncovers common problems in AI-enhanced educational secondary schools and targets alternative principles, which are demonstrated through examples that develop students’ component competencies, gain experience in constructing knowledge-centered logics of agency, and better apply intelligent software to assist learning. Literature [29] explores AI education workshops, expert reflections on the ethics of AI curricula, education on the topic of AI education at the early undergraduate secondary school level, and answers to the question of how AI can have a broad impact on multiple disciplines, deepening the knowledge and understanding of the role of AI in the field of education. Artificial Intelligence has brought disruptive changes to education, and experts have reflected on the relationship between AI and education through literature review methodology, thematic discussion sessions on personalized learning, intelligent educational data analysis, and AI educational interactions, etc., with the discussion and research on ethics in AI education being more popular.

In this paper, Generative Adversarial Networks are introduced to construct an instructional model based on AI mapping techniques. The architecture of GANs is composed of a generator and a discriminator. The generator receives a high-dimensional random noise vector as input, and through a series of transpose-convolution operations, the input low-dimensional noise vector is gradually up-sampled and transformed into a feature map with higher spatial dimensions. The discriminator is continuously trained against the generator to improve their own discriminative ability in order to more accurately recognize the graphs produced by the generator from the real design graph samples.The model is constructed and tested on the dataset for graph generation, and applied to teaching practice to analyze the effect.

2

Method

2.1

An innovative approach to teaching Graphic Design based on AI

2.1.1

Problems with traditional teaching models

The traditional teaching mode has a number of problems in the design profession, which limit the development of students’ comprehensive literacy and professional abilities.The traditional teaching mode pays too much attention to instilling knowledge and neglects the cultivation of practical skills and abilities. Students may have gained a certain amount of theoretical knowledge in the classroom, but they lack sufficient skills to support them in the actual design practice. The traditional teaching mode often lacks the opportunity to cultivate creative thinking and innovation. The field of design requires not only the use of technology, but also unique creativity and visual expression.However, the overly standardized teaching in the traditional mode often makes it difficult for students to fully utilize their personal unique style and imagination in design.

2.1.2

Innovative concept of teaching model based on AI mapping technology

The innovative teaching method that uses AI drawing technology aims to organically combine artificial intelligence technology with design education to offer students a more creative and practical learning experience. The teaching mode will change from a traditional teacher-centered mode to a student-centered mode, and the AI mapping technology can provide students with personalized learning and guidance, automatically adjusting the content and difficulty of the teaching according to the students’ interests, learning styles, and levels, so as to better meet the needs of each student.AI technology can collaborate with students to generate creative images, thus expanding their creative possibilities, and the teaching model can integrate design with a number of disciplines, including computer science and psychology, to produce more well-rounded and versatile design talent.

2.1.3

AI-based personalized learning and guidance

AI-based personalized learning and instruction is an educational approach dedicated to optimizing the student learning experience by personalizing and intelligently supporting the educational process through artificial intelligence technology. This approach aims to fully utilize the potential of each student by providing learning paths and teaching content tailored to their learning needs and interests. Personalized learning and guidance is based on AI technology, which customizes the learning plan for each student by analyzing information about the student’s learning history, study habits, interests, etc. AI algorithms are able to intelligently recommend textbooks, exercises, and projects that are appropriate for the student’s level based on their knowledge base and learning progress.Through this personalized learning approach, students can improve their skills in a more targeted way and improve their learning efficiency.

2.1.4

The Value of Generative Adversarial Networks in Design Instruction

The design of Generative Adversarial Networks (GANs) can be visualized as a two-player race in design creation: one is a skilled mimic and the other is a discerning connoisseur. The goal of mimicry is to create entirely new designs that are so visually indistinguishable from the real ones that even the most astute observer would have difficulty distinguishing them. In order to do this, the mimic uses a special technique of creating a piece from a starting point that is full of randomness, gradually adding details until the piece looks like it was made by a designer [30].

Instead, the connoisseur’s task is to identify which pieces are genuine designs and which were created by imitators. The connoisseur examines the details of the work to detect any potential flaws or inconsistencies. In this process, the two designers compete with each other to improve. Over time, the imitator improves his skills and creates more and more realistic pieces, while the connoisseur becomes sharper and is able to pick up on even smaller differences.Eventually, the imitator reaches a level where even the connoisseur has difficulty distinguishing the difference between his work and the real design.In this design-creation competition, the imitator is the equivalent of the “generator” in GANs, while the connoisseur is the equivalent of the “discriminator”.Through this continuous “confrontation”, the generator learns how to create new works of high quality and fidelity, while the discriminator improves its ability to evaluate and perceive the level of the work, as well as its own understanding of the design.

And in design education, students can observe, compare, analyze, and understand the elements and levels of design style, color, structure, details, emotional expression, and so on.Through the use of works generated against the network, the aim is to improve the efficiency of learning and teaching, and realize personalized design practices.

2.2

Graph design model based on generative adversarial networks

2.2.1

Self-attention mechanisms

Self-attention is a mechanism that has been developed in natural language processing and is described as a nonlocal operation in computer vision. It represents the features at each position in the feature map as a weighted sum of all the positional features, which makes a connection between the positional features and compensates for the limitation of the convolutional local receptive field.

For a feature map F with input shape [N,C,H,W], where N,C,H,W is the data batch, the number of feature map channels, the feature map height and the feature map width, respectively, the nonlocal operation is summarized in the following equation: 1 $F_{t, ∴ a, b}^{'} = \sum_{\int, d} \frac{\exp (Φ (F_{t, ∴ a, b}, F_{t, ∴ c, d}))}{\sum_{\int, f} \exp (Φ (F_{t, ∴ a, b}, F_{t, ∴ c, f}))} Ψ (F_{t, ∴ c, d})$

Where t,a,b denotes an index in the batch, height, and width dimensions of the feature map, respectively, and the “colon” denotes the entire channel dimension. The weight of the weighted sum is expressed as the match of the feature with all features using softmax activation, as indicated in Eq. In the model implementation, the transformation function ψ,Φ is represented as a 1×1-convolution halving the number of channels.

2.2.2

HSV color space

HSV/B divides colors into three channels: hue, saturation, and lightness. Unlike RGB, HSV provides a more intuitive representation of color by isolating luminance details from the other two channels so that the HS channel can differentiate between the outlines of objects that will be a single hue.The conversion from the RGB channel to the HSV channel is expressed as follows: 2 $h = {\begin{matrix} 0 & max = min \\ 60 \times \frac{g - b}{\max - \min} + 0 & \max = r a n d g \geq b \\ 60 \times \frac{g - b}{\max - \min} + 360 & \max = r a n d g < b \\ 60 \times \frac{b - r}{\max - \min} + 120 & \max = g \\ 60 \times \frac{r - g}{\max - \min} + 240 & \max = b \end{matrix}$ 3 $s = {\begin{matrix} 0 & \max = 0 \\ 1 - \frac{\min}{\max} & e l s e \end{matrix}$ 4 $v = \max$

where max, min denote the maximum and minimum of the RGB values, respectively. Since the conversion of RGB to HSV space involves selecting the maximum RGB value, which cannot be expressed by convolution, the convolution-based discriminator does not learn this pattern. Therefore, in this paper, we use the HS channel of the image spliced with the RGB channel as the image input to the discriminator so that the GAN model can better capture the shape structure of the objects in the image. The method only adds two layers of parameters to the convolution kernel of the discriminator input layer, which is a negligible increase in the number of parameters compared to the parameters of the entire model. Nonetheless, ablation experiments demonstrate the enhancement of the model’s representational capabilities.

2.2.3

Overall structure

In this paper, a residual attention multichannel generative adversarial network is designed to generate images with complex geometric structures.The model is divided into two parts: generator and discriminator. The basic structure is shown in Fig. 1. In order to improve the information transfer efficiency between the generator and the discriminator, this paper utilizes the jump connection and the residual connection structure to form the backbone network of the generator and the discriminator, respectively. The following describes the data transmission process of the generator and the discriminator, respectively [31].

1)

Generator

Suppose the latent variable is z ∈ R^N,C_ò,1,1, where N is the batch size and C_z is the number of channels of the latent variable. The intermediate feature maps and initial images are first obtained after the generator initialization module: 5 $F_{1} = O p s_{1} (z_{0})$ 6 $I_{1} = C o n v (F_{1})$

where Ops₁ is, in order, a transposed convolution of augmented feature map size, and a convolution with kernel size 3×3, respectively. I₁ is the initial image, which serves as the base image for later jump connections. After that, the features and images are fed into the various basic modules of the generator, expressed as follows: 7 $F_{G i} = O p s_{2} (F_{G i - 1}), i = 2, 3, \dots$ 8 $I_{i} = U p s a m p l e (I_{i - 1}) + C o n v (F_{G i}), i = 2, 3, \dots$

The Ops₂ operations are, in order, two-fold upsampling, 3×3 convolution that halves the number of channels, non-local operations, and 3×3 convolution that leaves the number of channels unchanged. The channels are F_Gi mapped to 3 by 1×1 convolution before performing elemental summation with I_i. After n basic blocks, the generator finally outputs an RGB channel image of size [2ⁿ⁺²,2ⁿ⁺²].

2)

Discriminator

For an RGB image tensor I ∈ R^N,3,2ⁿ⁺²,2ⁿ⁺² containing N image of dimension [2ⁿ⁺²,2ⁿ⁺²], the discriminator first computes its HS channels and splices them along the channel dimension to obtain the feature map F_D0 ∈ R^{N,5,2ⁿ⁺²,2ⁿ⁺²} 9 $F_{D 0} = C o n c a t e n a t e (I, H S (I))$

The feature map is then expressed as follows through the individual discriminator basic blocks: 10 $F_{D i} = O p s_{3} (F_{D i - 1}) + O p s_{4} (F_{D i - 1})$

In Ops₃, there is a 3×3-convolution to increase the number of channels, a non-local operation, a 3×3-convolution to keep the number of channels constant, and a two-fold downsampling, in that order. In Ops₄, a 1×1-convolution and a two-fold downsampling are used to vary the number of channels and resolution of the feature map for elemental summation. After n+1 basic blocks, the feature map F_Dn+1 is shaped as [N,2ⁿC_D,2,2]. Finally, the ending block of the discriminator is entered to obtain the discriminator result: 11 $P = DEndBlock(F_{D n + 1})$

In DEndBlock, the feature map undergoes a 3×3-convolution to amortize the last three dimensions into one dimension, which is finally mapped to a discriminant scalar through a linear layer.

2.2.4

Loss function

In order to prevent the discriminator from over-training resulting in gradient vanishing, in this paper, we use the hinge loss function as the GAN loss and the gradient penalty as the regular term of the discriminator parameters [32], which are expressed as follows: 12 $L_{G} = D (\begin{matrix} G (z_{i}) \end{matrix})$ 13 $L_{D} = Max(1 - D (G (z_{i})), 0) + Max(1 + D (x_{i}), 0)$ 14 $R_{D} = {(| | \nabla_{\hat{x}} D_{ω} (\hat{x}) | |_{2} - 1)}^{2}, w h e r e \hat{x} = ϵ G (z_{i}) + (1 - ϵ) x_{i}$

The objective functions of the generator and discriminator are respectively: 15 $O_{G} = \frac{1}{N} \sum_{i = 1}^{N} L_{G}$ 16 $O_{D} = \frac{1}{N} \sum_{i = 1}^{N} (L_{D} + λ R_{D})$

where z_i denotes latent variable sampling, x_i denotes true distribution sampling, ò is sampling of random variables obeying a uniform distribution of [0, 1], and λ is a constant for weighing regular terms, which is set to 10 in this paper.

2.3

Instructional model development based on AI mapping technology

2.3.1

Techniques and frameworks for the development of pedagogical tools

In the development of teaching tools based on AI mapping technology, the choice of technology and the design of the framework play a crucial role. In order to develop efficient, stable and feature-rich teaching tools, multiple technology elements and frameworks need to be considered comprehensively. For image generation and style transformation functions, deep learning frameworks such as TensorFlow, PyTorch, or Keras can be used, which provide powerful neural network construction and training functions and can support the development of complex models such as Generative Adversarial Networks (GANs).

For automated layout and design techniques, consider using a combination of front-end and back-end technologies. The front-end part can choose popular web development frameworks such as React, Vue.js, or Angular for building user-friendly interfaces that enable users to input and operate easily. The back-end can choose a language such as Python, combined with a corresponding web framework such as Django or Flask, for processing user requests, invoking the AI model and generating the corresponding design typography. The front-end user interface design is responsible for interacting with users and providing a friendly interface; the back-end server and API are responsible for handling user requests, communicating with AI models and processing data; the AI model integration and invocation part incorporates AI technology into the tool; and the data storage and management is realized through cloud services to ensure the scalability and stability of the tool.

2.3.2

Design and Implementation of Creative Image Generation Aids

The design and implementation of the Creative Image Generation Aid involve several technical aspects, aiming to provide designers with a powerful tool that can stimulate creativity and aid in creation. In the design and implementation, a suitable deep learning framework, such as TensorFlow or PyTorch, needs to be selected for building the generation model. The architecture and parameters of the model need to be carefully selected and adjusted to obtain high-quality generative results. In addition, a large image dataset needs to be prepared for model training to ensure that the model learns diverse creative styles and elements. In order for designers to interact with the tool, user-friendly interfaces need to be developed. Web development frameworks, such as React or Vue.js, can be chosen to build an intuitive front-end interface in which designers can enter initial conditions, select styles, adjust parameters, and so on. By communicating with the back-end server, the front-end can pass the user’s choices to the generative model and feed the generated image back to the user. On the back-end side, languages such as Python combined with web frameworks such as Django or Flask can be used to handle front-end requests and interactions with the model. The back-end needs to be responsible for passing the designer’s input to the generative model, processing the images generated by the model, and returning the results to the front-end for display. At the same time, the backend needs to ensure data security and efficient operation of the model.

3

Results and Discussion

3.1

Test of model generation results

3.1.1

Residual connection validity

An experiment is first designed to verify the validity of the model in this paper for residual connections. The objective of this experiment is to examine how the residual connection affects the MMD scores of the generated graphs when it is employed in the discriminator. In terms of experimental design, this section chooses to demonstrate its effectiveness on the ENZYMES dataset by constructing different neural network parameters of the discriminator, the aggregation function of the last layer of the residual connection, and whether or not to use the residual connection in order to prove its effectiveness, and to help the model proposed in this chapter to choose the most suitable model parameters.Among them, the baseline models are MolGAN and DCGAN approaches.The MolGAN model is similar to the one presented in this chapter, but lacks a residual linking mechanism, so it can be limited by having a discriminator that is too shallow.DCGAN models use residuals. The models have similar constraints.

Figures 2 to 4 show the comparison of MMD score results for each model. Among them, Fig. 2 shows the comparison of node degree distribution, Fig. 3 shows the comparison of clustering coefficient distribution, and Fig. 4 shows the comparison of average number of orbits distribution.

It can be seen that the model introduced in this paper can help to mitigate the effect of the vanishing gradient problem in deep graph convolutional networks by introducing residual connections. As can be seen from the result graphs, the performance of the MolGAN model improves only in two or three graph neural network layers, and there is no significant performance improvement after more than three. In contrast, the discriminator construction of the GBENWEN model structure, due to the inclusion of residual connections, can still show performance improvement after more than three layers, and even performs 89.5% less on the number of orbits distribution MMD performance metrics than the case without residual connections. Except for the clustering coefficient distribution MMD metric, where the advantage of residual connectivity is not as significant, overall, increasing the number of graph neural network layers and using residual connectivity is very effective.

3.1.2

Generating results tests

Table 1 shows the results of this paper’s model compared with the baseline model on the designed text dataset, containing the MMD values between the node degree distribution, cluster coefficient distribution, and mean track count distribution and the real data. Among them, the smaller the MMD value, the better. From the table, it can be observed that traditional graph generation models do not perform as well as graph generation models utilizing deep learning algorithms on complex graph datasets in general. Such a result is understandable because the basis of ER and BA model generation is based on mathematical models that do not fit well with the graph features in different types of complex datasets.The model developed in this paper exhibits a significant improvement and a significant decrease in MMD values compared to the traditional baseline algorithms for graph generation.

Table 1.

Compare the generated results in the design text data set

	Cora			Citeseer
	Degree	Clustering	Orbit	Degree	Clustering	Orbit
ER	0.74	0.86	0.46	0.67	0.68	0.14
BA	0.31	0.62	0.35	0.36	0.44	0.16
GraphRNN	0.33	0.47	0.19	0.28	0.35	0.11
MolGAN	0.23	0.26	0.25	0.22	0.21	0.37
This model	0.18	0.19	0.08	0.18	0.19	0.06

The graphical dataset shows the results of this paper’s model compared to the baseline model in Table 2. It can be seen that the MMD of this paper’s model is the lowest among all models in all dimensions.

Table 2.

Compare the generated results in the Graphic data set

	PROTEINS			ENZYMES
	Degree	Clustering	Orbit	Degree	Clustering	Orbit
ER	0.64	1.02	0.24	0.37	1.34	0.08
BA	0.96	0.84	0.08	1.22	1.01	0.26
GraphRNN	0.75	0.62	0.14	0.45	0.36	0.09
MolGAN	0.15	0.17	0.16	0.08	0.21	0.12
This model	0.11	0.14	0.09	0.19	0.19	0.04

3.2

Analysis of the results of the experiment on teaching Graphic Design

In order to test the application effect of the model constructed in this paper in the classroom of design majors, the model is applied to the design majors of a university, and two classes are selected as the experimental group and the control group to carry out the teaching experiment of the course “Graphic Design”. The experimental group utilizes the model for teaching and self-study activities, while the control group teaches according to the original model. An examination was conducted at the end to test the learning outcomes.

The examination test is 12 items, which are (1) color theory, (2) layout design, (3) typography, (4) visual hierarchy, (5) image representation, (6) UI design, (7) visual experience management, (8) image element basics, (9) graphic editing, (10) 3D modeling and rendering, (11) use of software tools, and (12) kinetic design. Each item is worth 5 points.

At the end of the course implementation, independent t-tests were conducted to compare the differences between the test results of the two groups. The results of the mean scores are shown in Figure 5. The average score of the experimental group was 50.65, which was 7.28 points higher than the 43.37 score of the control group, and it was ahead of the control group on all 12 items examined.

Table 3 shows the results of the independent samples t-test for performance. The Sig (two-tailed value) is less than 0.05 for all questions except for questions 3, 4 and 5 of the scale, indicating that there is a significant difference between the experimental group and the control group in these items, while there is no significant difference in items 3, 4 and 5. Overall, the application of the model and methodology presented in this paper is beneficial in designing professional classroom teaching.

Table 3.

Independent sample t test

Independent sample inspection
N		F	Sig.	t	df	Sig.(Double)
1	Assumed equal variance	0.002	0.985	4.037	100	0.000
1	Unassuming equal variance	0.002	0.985	4.037	81.95	0.000
2	Assumed equal variance	3.394	0.066	3.025	100	0.004
2	Unassuming equal variance	3.394	0.066	3.025	95.04	0.004
3	Assumed equal variance	1.872	0.178	-0.447	100	0.650
3	Unassuming equal variance	1.872	0.178	-0.447	95.79	0.650
4	Assumed equal variance	0.563	0.459	0	100	1.020
4	Unassuming equal variance	0.563	0.459	0	84.76	1.020
5	Assumed equal variance	15.816	0	1.808	100	0.078
5	Unassuming equal variance	15.816	0	1.808	85.3	0.078
6	Assumed equal variance	0.695	0.409	4.033	100	0.000
6	Unassuming equal variance	0.695	0.409	4.033	94.85	0.000
7	Assumed equal variance	22.846	0	2.918	100	0.006
7	Unassuming equal variance	22.846	0	2.918	91.12	0.006
8	Assumed equal variance	4.685	0.032	7.256	100	0.000
8	Unassuming equal variance	4.685	0.032	7.256	86.25	0.000
9	Assumed equal variance	3.916	0.065	2.451	100	0.017
9	Unassuming equal variance	3.916	0.065	2.451	84.59	0.017
10	Assumed equal variance	3.628	0.058	5.078	100	0.000
10	Unassuming equal variance	3.628	0.058	5.078	91.75	0.000
11	Assumed equal variance	2.574	0.107	2.336	100	0.025
11	Unassuming equal variance	2.574	0.107	2.336	83.67	0.025
12	Assumed equal variance	0.038	0.857	3.874	100	0.000
12	Unassuming equal variance	0.038	0.857	3.874	86.16	0.000

3.3

Optimization of graphic design teaching by artificial intelligence technology

3.3.1

The advantage of using AI technology in graphic design teaching

Through the teaching experiment results of Graphic Design, it is found that the advantages of using AI technology in graphic design teaching are as follows:

1) The application of AI technology in graphic design teaching is reflected in the expansion of graphic design teaching content.

Due to the development of artificial intelligence technology, AI technology is no longer limited to image processing and graphic design, but is integrated with graphic design, and this is used to develop new course content. For example, in the course "Advertising Creativity", new course content has been developed by using AI technology. These course contents are difficult to carry out in traditional teaching, but they can become easier from the perspective of artificial intelligence.

2) It can cultivate students’ innovative consciousness and creative ability.

Graphic design teaching is a course that focuses on cultivating students’ innovative consciousness and ability. Only when students have innovative abilities can they provide valuable services to enterprises. However, in traditional graphic design teaching, both teachers and students are used to learning graphic design knowledge from books and theories. AI technology has the ability to analyze and process a large number of image files and turn them into data that AI algorithms can recognize and analyze.After analyzing these data using an AI algorithm, text, images, animations, and other content can be generated, and these contents can meet the creative and innovative requirements of graphic design teaching.

3) Can effectively improve the teaching efficiency of graphic design.

In traditional graphic design teaching, teachers spend a lot of time explaining theory and demonstrating in the classroom to students, but the effect is not satisfactory.After using AI technology to develop new course content, teachers can directly turn these course contents into courseware for demonstration in class, and students’ understanding and mastery of these materials will be more efficient.

3.3.2

Innovative teaching strategies of graphic design in universities

1) Use artificial intelligence image technology to promote personalized graphic design teaching.

In order to integrate graphic design teaching and artificial intelligence image technology in colleges and universities, stimulate students’ personalized creative ability, and then achieve the purpose of transformation and upgrading, teachers need to make full use of artificial intelligence image technology to expand teaching content and promote the establishment of personalized graphic design teaching objectives.

2) Develop artificial intelligence image courses to improve visual graphic design teaching.

The integration of artificial intelligence image technology makes the traditional graphic design teaching in colleges and universities no longer rely on computer graphic software in a single way, but through intelligent AI design platform or DALL-E tool system. Therefore, teachers should actively develop artificial intelligence image courses and improve the teaching methods for visual graphic design.

3) Expand artificial intelligence algorithm learning and innovate new forms of graphic design teaching.

Intelligent algorithms are an indispensable learning component of artificial intelligence image technology. Although it is difficult for design students to learn, teachers still need to expand the learning content of artificial intelligence algorithm in course learning, and innovate new forms of graphic design teaching according to students’ classroom learning.

4

Conclusion

This paper focuses on the pedagogical application of GANs technology for graph design and generation. A pedagogical model based on artificial intelligence (AI) graphics technology has been developed.The model generation performance was analyzed on the dataset and it was found that the MMD of this paper’s model is the lowest among all models on all dimensions of the graphics dataset. The introduction of the attention mechanism and the use of residual connections in this paper is very effective compared to the baseline model without adding residual connections. In the teaching experiment, the average score of the experimental group is 50.65, which is 16.79% higher than that of the control group, and the scores of all 12 items examined are above 4. The model and method of this paper can significantly improve learning and teaching efficiency, and its effectiveness has been confirmed in teaching practice.

Język:: Angielski

Częstotliwość wydawania:: 1 razy w roku
Dziedziny czasopisma:: Nauki biologiczne, Nauki biologiczne, inne, Matematyka, Matematyka stosowana, Matematyka ogólna, Fizyka, Fizyka, inne

Kanał RSS czasopisma

Exploring the Application of Artificial Intelligence in Design Courses in Colleges and Universities: an Example from Graphic Design

Minggang Liu

Jinding Zou

Data publikacji: 21 mar 2025

Otrzymano: 25 paź 2024

Przyjęty: 18 lut 2025

DOI: https://doi.org/10.2478/amns-2025-0651

Słowa kluczoweGenerative adversarial network, Residual connection, Attention mechanism, Graphic design, Design professional teaching

© 2025 Minggang Liu et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Słowa kluczowe
Generative adversarial network, Residual connection, Attention mechanism, Graphic design, Design professional teaching